
from ecg_qc import ecg_qc
import time
import numpy as np
import matplotlib.pyplot as plt
from joblib import load
from dtreeviz.trees import *
import pandas as pd
from sklearn import tree
from streamlit_visualization.modules.graph_generation import annot_classification_correspondance
import seaborn as sns
from sklearn.metrics import f1_score, accuracy_score
from sklearn.tree import DecisionTreeClassifier
from sklearn.model_selection import GridSearchCV
from sklearn.preprocessing import StandardScaler
from sklearn.cluster import KMeans


df_train_sqi = pd.read_csv('training_dataset/df_consolidated_ml.csv', index_col=0)
df_train_sqi.head()


# Val Dataset
df_val_sqi = pd.read_csv('validation_dataset/df_ml_103001_2s.csv')
df_val_sqi.head()


### Initialisation of ecg_qc instances

ecg_qc_rfc = ecg_qc(model='training_dataset/model_2s_rfc_normalized_premium.pkl',
                      normalized=True)
#ecg_qc_rfc = ecg_qc(normalized=False)
ecg_qc_dtc = ecg_qc(model='training_dataset/model_dtc_direct.pkl',
                     normalized=False)

try:
    ecg_qc_rfc.model = ecg_qc_rfc.model.best_estimator_
except:
    pass

/home/aura-alexis/github/ecg_qc_viz/env/lib64/python3.6/site-packages/scikit_learn-0.24.1-py3.6-linux-x86_64.egg/sklearn/base.py:315: UserWarning:

Trying to unpickle estimator StandardScaler from version 0.23.2 when using version 0.24.1. This might lead to breaking code or invalid results. Use at your own risk.

/home/aura-alexis/github/ecg_qc_viz/env/lib64/python3.6/site-packages/scikit_learn-0.24.1-py3.6-linux-x86_64.egg/sklearn/base.py:315: UserWarning:

Trying to unpickle estimator OneHotEncoder from version 0.23.2 when using version 0.24.1. This might lead to breaking code or invalid results. Use at your own risk.

/home/aura-alexis/github/ecg_qc_viz/env/lib64/python3.6/site-packages/scikit_learn-0.24.1-py3.6-linux-x86_64.egg/sklearn/base.py:315: UserWarning:

Trying to unpickle estimator ColumnTransformer from version 0.23.2 when using version 0.24.1. This might lead to breaking code or invalid results. Use at your own risk.

/home/aura-alexis/github/ecg_qc_viz/env/lib64/python3.6/site-packages/scikit_learn-0.24.1-py3.6-linux-x86_64.egg/sklearn/base.py:315: UserWarning:

Trying to unpickle estimator DecisionTreeClassifier from version 0.24.2 when using version 0.24.1. This might lead to breaking code or invalid results. Use at your own risk.

/home/aura-alexis/github/ecg_qc_viz/env/lib64/python3.6/site-packages/scikit_learn-0.24.1-py3.6-linux-x86_64.egg/sklearn/base.py:315: UserWarning:

Trying to unpickle estimator RandomForestClassifier from version 0.24.2 when using version 0.24.1. This might lead to breaking code or invalid results. Use at your own risk.

/home/aura-alexis/github/ecg_qc_viz/env/lib64/python3.6/site-packages/scikit_learn-0.24.1-py3.6-linux-x86_64.egg/sklearn/base.py:315: UserWarning:

Trying to unpickle estimator GridSearchCV from version 0.24.2 when using version 0.24.1. This might lead to breaking code or invalid results. Use at your own risk.


# Adding Predictions

for dataframe in [df_train_sqi, df_val_sqi]:

    dataframe['rfc_pred'] = [ecg_qc_rfc.predict_quality([x]) for x in dataframe.iloc[:, 2:8].values]
    dataframe['dtc_pred'] = [ecg_qc_dtc.predict_quality([x]) for x in dataframe.iloc[:, 2:8].values]
    
df_val_sqi['consensus'] = df_val_sqi['classif']


# Compute score for both dataset and both models

for dataframe in [df_train_sqi, df_val_sqi]:
    for col in ['rfc_pred', 'dtc_pred']:

        print(f'{col}:')
        print(f"f1_score : {f1_score(dataframe[col].values, dataframe['consensus'].values)}")
        print(f"accuracy_score: {accuracy_score(dataframe[col].values, dataframe['consensus'].values)}")
        print(dataframe['consensus'].value_counts()/dataframe.shape[0])
        print('\n')

rfc_pred:
f1_score : 0.7920771965464702
accuracy_score: 0.6898484848484848
1    0.593333
0    0.406667
Name: consensus, dtype: float64


dtc_pred:
f1_score : 0.9676610679368264
accuracy_score: 0.9609090909090909
1    0.593333
0    0.406667
Name: consensus, dtype: float64


rfc_pred:
f1_score : 0.676300578034682
accuracy_score: 0.5333333333333333
1    0.6625
0    0.3375
Name: consensus, dtype: float64


dtc_pred:
f1_score : 0.6263020833333334
accuracy_score: 0.5216666666666666
1    0.6625
0    0.3375
Name: consensus, dtype: float64


df_train_sqi.iloc[:,2:8].describe()


df_val_sqi.iloc[:,2:8].describe()


df_train_sqi.iloc[:,2:8].describe() - df_val_sqi.iloc[:,2:8].describe()


# Working cases
df_train_ = df_train_sqi[df_train_sqi['consensus'] == df_train_sqi['dtc_pred']]
df_val_ = df_val_sqi[df_val_sqi['consensus'] == df_val_sqi['dtc_pred']]

df_train_.iloc[:,2:8].describe() - df_val_.iloc[:,2:8].describe()


df_train_.iloc[:,2:8].describe()


df_val_.iloc[:,2:8].describe()


# sSQi and kSqi are issues


# Global - working
df_train_sqi.iloc[:,2:8].describe() - df_val_sqi.iloc[:,2:8].describe() - (df_train_.iloc[:,2:8].describe() - df_val_.iloc[:,2:8].describe())


# Not Working cases
df_train_ = df_train_sqi[df_train_sqi['consensus'] != df_train_sqi['dtc_pred']]
df_val_ = df_val_sqi[df_val_sqi['consensus'] != df_val_sqi['dtc_pred']]

df_train_.iloc[:,2:8].describe() - df_val_.iloc[:,2:8].describe()


df_train_.iloc[:,2:8].describe()


df_val_.iloc[:,2:8].describe()


# Global - not Working
df_train_sqi.iloc[:,2:8].describe() - df_val_sqi.iloc[:,2:8].describe() - (df_train_.iloc[:,2:8].describe() - df_val_.iloc[:,2:8].describe())


# Trying with simpler elements

df_simple = df_train_sqi.loc[:, ['qSQI_score', 'cSQI_score', 'pSQI_score', 'basSQI_score']]
X_simple = df_simple.values
y = df_train_sqi['consensus'].tolist()
model = DecisionTreeClassifier()

algo = DecisionTreeClassifier()
params = {'min_samples_leaf': np.arange(1, 5, 1),
      'max_depth': np.arange(2, 25, 1),
      'max_features': [4]
      }
grid_search = GridSearchCV(estimator=algo,
                       param_grid=params,
                       scoring='f1',
                       cv=5,
                       verbose=5,
                       n_jobs=-1)

grid_search.fit(X_simple, y)
y_pred = grid_search.predict(X_simple)
print(f'f1_score : {f1_score(y, y_pred)}')
print(f'accuracy_score : {accuracy_score(y, y_pred)}')

print('\n val')
X_val = df_val_sqi.loc[:, ['qSQI_score', 'cSQI_score', 'pSQI_score', 'basSQI_score']].values
y_val = df_val_sqi['consensus'].values
y_pred_val = grid_search.predict(X_val)
print(f'f1_score : {f1_score(y_val, y_pred_val)}')
print(f'accuracy_score : {accuracy_score(y_val, y_pred_val)}')

Fitting 5 folds for each of 92 candidates, totalling 460 fits
f1_score : 0.8814753905791143
accuracy_score : 0.8471212121212122

 val
f1_score : 0.5850622406639004
accuracy_score : 0.5


# algo = RandomForestClassifier()
algo = DecisionTreeClassifier()

params = {'min_samples_leaf': np.arange(1, 5, 1),
      'max_depth': np.arange(2, 16, 2),
      'max_features': ['auto'],
      }
grid_search = GridSearchCV(estimator=algo,
                       param_grid=params,
                       scoring='f1',
                       cv=5,
                       verbose=5,
                       n_jobs=-1)
grid_search.fit(X_train, y_train)


forest = ecg_qc_rfc.model

sqi_names = ['q_sqi_score', 'c_sqi_score', 's_sqi_score',
             'k_sqi_score', 'p_sqi_score', 'bas_sqi_score']
feature_names = sqi_names


# Importance des features

importances = forest.feature_importances_
std = np.std([
    tree.feature_importances_ for tree in forest.estimators_], axis=0)

forest_importances = pd.Series(importances, index=feature_names)

fig, ax = plt.subplots()
forest_importances.plot.bar(yerr=std, ax=ax)
ax.set_title("Feature importances using MDI")
ax.set_ylabel("Mean decrease in impurity")
fig.tight_layout()


# Importance des features

importances = ecg_qc_dtc.model.feature_importances_
forest_importances = pd.Series(importances, index=feature_names)

fig, ax = plt.subplots()
forest_importances.plot.bar(ax=ax)
ax.set_title("Feature importances using MDI")
ax.set_ylabel("Mean decrease in impurity")
fig.tight_layout()


# Train dataset

dataset = df_train_sqi
model = ecg_qc_dtc.model

X = dataset.iloc[:, 2:8].values
y_pred = dataset['dtc_pred'].tolist()
y_true = np.array(dataset['consensus'].values)

print(f'True :{dataset["consensus"].values[0]}')
print(f'Pred :{dataset["dtc_pred"].values[0]}')


viz = dtreeviz(model, X, y_true,
            target_name='quality',
            feature_names=feature_names,
            class_names=['noise', 'clean'],
            X=X[0],
            show_just_path=True)

viz

True :1
Pred :1


# Val dataset

dataset = df_val_sqi
model = ecg_qc_dtc.model

X = dataset.iloc[:, 2:8].values
y_pred = dataset['dtc_pred'].tolist()
y_true = np.array(dataset['consensus'].values)

print(f'True :{dataset["consensus"].values[0]}')
print(f'Pred :{dataset["dtc_pred"].values[0]}')


viz = dtreeviz(model, X, y_true,
            target_name='quality',
            feature_names=feature_names,
            class_names=['noise', 'clean'])

viz

True :1
Pred :1


# Val dataset

dataset = df_val_sqi[df_val_sqi['consensus'] != df_val_sqi['dtc_pred']]
model = ecg_qc_dtc.model

X = dataset.iloc[:, 2:8].values
y_pred = dataset['dtc_pred'].tolist()
y_true = np.array(dataset['consensus'].values)

print(f'True :{dataset["consensus"].values[0]}')
print(f'Pred :{dataset["dtc_pred"].values[0]}')


viz = dtreeviz(model, X, y_true,
            target_name='quality',
            feature_names=feature_names,
            class_names=['noise', 'clean'],
            X=X[0],
            show_just_path=True)

viz

True :1
Pred :0


ct = ctreeviz_bivar(ecg_qc_dtc.model, df_val_sqi.iloc[:, 2:8], df_val_sqi['consensus'],
                    feature_names = ['PEG','LPR'],
                    class_names=['noisy', 'clean'],
                    target_name='Knowledge')
plt.tight_layout()
plt.show()


# Decision Path

print(f'True :{df_wrong_answers["classif"].values[0]}')
print(f'False :{df_wrong_answers["ecg_qc_pred"].values[0]}')

viz = dtreeviz(forest.estimators_[0], X, y_true,
                target_name='quality',
                feature_names=feature_names,
                class_names=['noise', 'clean'],
                X=X[0],
                show_just_path=True)

viz


X = df_val_sqi.iloc[:, 2:8].values
y_pred = df_val_sqi['rfc_pred'].tolist()
y_true = np.array(df_val_sqi['consensus'].values)


viz = dtreeviz(forest.estimators_[0], X, y_true,
                target_name='quality',
                feature_names=feature_names,
                class_names=['noise', 'clean'])

viz.save('viz_tree_103001.svg')

2021-08-02 14:59:29.191 WARNING matplotlib.font_manager: findfont: Font family ['Arial'] not found. Falling back to DejaVu Sans.
2021-08-02 14:59:29.207 WARNING matplotlib.font_manager: findfont: Font family ['Arial'] not found. Falling back to DejaVu Sans.
2021-08-02 15:00:39.256 WARNING matplotlib.font_manager: findfont: Font family ['Arial'] not found. Falling back to DejaVu Sans.


for kmean in range(5):
    print(f'kmeans {kmean}')

    df_temp = df_val_sqi[df_val_sqi['kmeans'] == kmean]
    X_temp = df_temp.iloc[:,2:8].values
    print(f'count : {df_temp.shape[0]}')
    print(df_temp['classif'].value_counts()/df_temp.shape[0])
    y_true_temp = np.array(df_temp['classif'].values)
    predictions = [ecg_qc_rfc.predict_quality([x]) for x in X_temp]
    print(f1_score(y_true_temp, predictions))
    print(accuracy_score(y_true_temp, predictions))
    dis = plt.figure()
    dis = sns.distplot(df_temp['qSQI_score'].values)
    print('\n')

kmeans 0
count : 448
1    0.53125
0    0.46875
Name: classif, dtype: float64
0.6204620462046204
0.48660714285714285

/home/aura-alexis/github/ecg_qc_viz/env/lib64/python3.6/site-packages/seaborn/distributions.py:2557: FutureWarning:

`distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).


kmeans 1
count : 142
1    0.950704
0    0.049296
Name: classif, dtype: float64
0.055172413793103454
0.035211267605633804


kmeans 2
count : 113
0    0.548673
1    0.451327
Name: classif, dtype: float64

/home/aura-alexis/github/ecg_qc_viz/env/lib64/python3.6/site-packages/seaborn/distributions.py:2557: FutureWarning:

`distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).

0.6134969325153374
0.4424778761061947


kmeans 3
count : 49
0    0.938776
1    0.061224
Name: classif, dtype: float64

/home/aura-alexis/github/ecg_qc_viz/env/lib64/python3.6/site-packages/seaborn/distributions.py:2557: FutureWarning:

`distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).

0.048780487804878044
0.20408163265306123


kmeans 4
count : 448
1    0.821429
0    0.178571
Name: classif, dtype: float64

/home/aura-alexis/github/ecg_qc_viz/env/lib64/python3.6/site-packages/seaborn/distributions.py:2557: FutureWarning:

`distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).

0.8825806451612903
0.796875

/home/aura-alexis/github/ecg_qc_viz/env/lib64/python3.6/site-packages/seaborn/distributions.py:2557: FutureWarning:

`distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).


df_validation[df_validation['kmeans'] == 0]


# Decision Path

print(f'True :{df_wrong_answers["classif"].values[0]}')
print(f'False :{df_wrong_answers["ecg_qc_pred"].values[0]}')

viz = dtreeviz(forest.estimators_[0], X, y_true,
                target_name='quality',
                feature_names=feature_names,
                class_names=['noise', 'clean'],
                X=X[0],
                show_just_path=True)

viz

True :0
False :1


# Decision Path


viz = dtreeviz(forest.estimators_[0], X, y_true,
                target_name='quality',
                feature_names=feature_names,
                class_names=['noise', 'clean'],
                X=X[0],
                show_just_path=True)

viz.save('viz_tree_103001_pred.svg')


from sklearn.metrics import f1_score, accuracy_score


f1_score(y_true, y_pred)

0.7969924812030075


accuracy_score(y_true, y_pred)

0.6625


df_validation['classif'].value_counts()/df_validation.shape[0]

1    0.6625
0    0.3375
Name: classif, dtype: float64


df_wrong_answers = df_validation[df_validation['classif'] != df_validation['ecg_qc_pred']]


df_wrong_answers.iloc[:,2:8].values

array([[ 0.8  ,  0.784,  3.175, 12.127,  0.558,  0.964],
       [ 0.   ,  0.765,  2.276,  7.542,  0.487,  0.889],
       [ 0.   ,  0.42 , -0.126,  1.498,  0.499,  0.734],
       ...,
       [ 0.75 ,  0.666,  3.085, 10.528,  0.518,  0.948],
       [ 0.333,  0.528,  2.802, 10.864,  0.527,  0.939],
       [ 0.4  ,  0.725,  3.564, 13.994,  0.501,  0.926]])


regr = tree.DecisionTreeClassifier(max_depth=10)
regr.fit(X, y_true)
print(f1_score(y_true, regr.predict(X)))
print(accuracy_score(y_true, regr.predict(X)))

0.9460154241645244
0.93


viz = dtreeviz(regr, X, y_true,
                target_name='quality',
                feature_names=feature_names,
                class_names=['noise', 'clean'])

viz


# Decision Path
n = 5
x_ex = df_wrong_answers.iloc[:,2:8].values[n]
print(f'True :{df_wrong_answers["classif"].values[n]}')
print(f'False :{regr.predict([x_ex])}')
viz = dtreeviz(regr, X, y_true,
                target_name='quality',
                feature_names=feature_names,
                class_names=['noise', 'clean'],
                X=df_wrong_answers.iloc[:,2:8].values[5],
                show_just_path=True)
viz

True :0
False :[0]


# Decision Path

print(f'True :{df_wrong_answers["classif"].values[0]}')
print(f'False :{df_wrong_answers["ecg_qc_pred"].values[0]}')
viz = dtreeviz(forest.estimators_[0], X, y_true,
                target_name='quality',
                feature_names=feature_names,
                class_names=['noise', 'clean'],
                X=df_wrong_answers.iloc[:,2:8].values[0],
                show_just_path=True)
viz.view()

True :0
False :1

viz


# Kmeans
# Import Standard Scaler
X = df_val_sqi.iloc[:,2:8].values

sc = StandardScaler()
X_kmeans = sc.fit_transform(X)
kmeans = KMeans(n_clusters=5, random_state=0, init="k-means++")
kmeans.fit(X_kmeans)


# Cluster centers coordinates
# Let's use inverse transform to have the real coordinates 
cluster_centers = sc.inverse_transform(kmeans.cluster_centers_)


# Create a heatmap 
sns.heatmap(pd.DataFrame(cluster_centers, columns=sqi_names),  annot=True)

<AxesSubplot:>


df_val_sqi['kmeans'] = kmeans.predict(X_kmeans)


wcss =  []
k = []
for i in range (1,11): 
    kmeans = KMeans(n_clusters= i, init = "k-means++", random_state = 0)
    kmeans.fit(X_kmeans)
    wcss.append(kmeans.inertia_)
    k.append(i)
    print("WCSS for K={} --> {}".format(i, wcss[-1]))

WCSS for K=1 --> 7200.000000000006
WCSS for K=2 --> 5553.648224134901
WCSS for K=3 --> 4590.04396829362
WCSS for K=4 --> 3945.2204822451235
WCSS for K=5 --> 3421.1051848885336
WCSS for K=6 --> 3136.9221752794765
WCSS for K=7 --> 2896.336555136839
WCSS for K=8 --> 2731.89591420912
WCSS for K=9 --> 2527.422402670769
WCSS for K=10 --> 2368.9552345306865


import plotly.express as px

# Create DataFrame
wcss_frame = pd.DataFrame(wcss)
k_frame = pd.Series(k)

# Create figure
fig= px.line(
    wcss_frame,
    x=k_frame,
    y=wcss_frame.iloc[:,-1]
)

# Create title and axis labels
fig.update_layout(
    yaxis_title="Inertia",
    xaxis_title="# Clusters",
    title="Inertia per cluster"
)

# Render
#fig.show(renderer="notebook")
fig.show(renderer="iframe") # if using workspace


df_val_sqi.columns

Index(['timestamp_start', 'timestamp_end', 'qSQI_score', 'cSQI_score',
       'sSQI_score', 'kSQI_score', 'pSQI_score', 'basSQI_score', 'classif',
       'classif_avg', 'rfc_pred', 'dtc_pred', 'consensus', 'kmeans'],
      dtype='object')


fig = px.scatter(df_val_sqi, x='sSQI_score', y='kSQI_score', color='kmeans')
fig.show()


for kmean in range(5):
    
    df_temp = df_val_sqi[df_val_sqi['kmeans'] == kmean]

    print(f'\n val for kmean {kmean}')
    X_val = df_temp.iloc[:, 2:8].values
    y_val = df_temp['consensus'].values
    y_pred_val = ecg_qc_dtc.model.predict(X_val)
    print(f'f1_score : {f1_score(y_val, y_pred_val)}')
    print(f'accuracy_score : {accuracy_score(y_val, y_pred_val)}')
    print(f'size : {len(y_val)}')
    print(f"% of quality: ': {df_temp['consensus'].mean()}")

 val for kmean 0
f1_score : 0.5080645161290323
accuracy_score : 0.45535714285714285
size : 448
% of quality: ': 0.53125

 val for kmean 1
f1_score : 0.4228571428571428
accuracy_score : 0.2887323943661972
size : 142
% of quality: ': 0.9507042253521126

 val for kmean 2
f1_score : 0.5131578947368421
accuracy_score : 0.34513274336283184
size : 113
% of quality: ': 0.45132743362831856

 val for kmean 3
f1_score : 0.06666666666666667
accuracy_score : 0.42857142857142855
size : 49
% of quality: ': 0.061224489795918366

 val for kmean 4
f1_score : 0.8140556368960468
accuracy_score : 0.7165178571428571
size : 448
% of quality: ': 0.8214285714285714


# apply kmeans from train model

# Kmeans
# Import Standard Scaler
X = df_train_sqi.iloc[:,2:8].values

sc = StandardScaler()
X_kmeans = sc.fit_transform(X)
kmeans = KMeans(n_clusters=5, random_state=0, init="k-means++")
kmeans.fit(X_kmeans)
df_train_sqi['kmeans'] = kmeans.predict(X_kmeans)

X_val = df_val_sqi.iloc[:,2:8].values
X_val = sc.transform(X_val)
cluster_centers = sc.inverse_transform(kmeans.cluster_centers_)

df_val_sqi['kmeans'] = kmeans.predict(X_val)


for kmean in range(5):
    
    df_temp = df_train_sqi[df_train_sqi['kmeans'] == kmean]

    print(f'\n val for kmean {kmean}')
    X_val = df_temp.iloc[:, 2:8].values
    y_val = df_temp['consensus'].values
    y_pred_val = ecg_qc_dtc.model.predict(X_val)
    print(f'f1_score : {f1_score(y_val, y_pred_val)}')
    print(f'accuracy_score : {accuracy_score(y_val, y_pred_val)}')
    print(f'size : {len(y_val)}')
    print(f'% if total : {len(y_val)/df_train_sqi.shape[0]}')
    print(f"% of quality: ': {df_temp['consensus'].mean()}")

 val for kmean 0
f1_score : 0.0
accuracy_score : 0.9981114258734656
size : 1059
% if total : 0.16045454545454546
% of quality: ': 0.0

 val for kmean 1
f1_score : 0.9651081646894626
accuracy_score : 0.9481865284974094
size : 1930
% if total : 0.2924242424242424
% of quality: ': 0.7248704663212435

 val for kmean 2
f1_score : 0.6549707602339182
accuracy_score : 0.8959435626102292
size : 567
% if total : 0.08590909090909091
% of quality: ': 0.13756613756613756

 val for kmean 3
f1_score : 0.9569288389513109
accuracy_score : 0.9437652811735942
size : 1636
% if total : 0.24787878787878787
% of quality: ': 0.6356968215158925

 val for kmean 4
f1_score : 0.9982161969318588
accuracy_score : 0.9964488636363636
size : 1408
% if total : 0.21333333333333335
% of quality: ': 0.9936079545454546


for kmean in range(5):
    
    df_temp = df_val_sqi[df_val_sqi['kmeans'] == kmean]

    print(f'\n val for kmean {kmean}')
    X_val = df_temp.iloc[:, 2:8].values
    y_val = df_temp['consensus'].values
    y_pred_val = ecg_qc_dtc.model.predict(X_val)
    print(f'f1_score : {f1_score(y_val, y_pred_val)}')
    print(f'accuracy_score : {accuracy_score(y_val, y_pred_val)}')
    print(f'size : {len(y_val)}')
    print(f'% if total : {len(y_val)/df_val_sqi.shape[0]}')
    print(f"% of quality: ': {df_temp['consensus'].mean()}")

 val for kmean 0
f1_score : 0.0
accuracy_score : 0.0
size : 4
% if total : 0.0033333333333333335
% of quality: ': 0.0

 val for kmean 1
f1_score : 0.806015037593985
accuracy_score : 0.6950354609929078
size : 423
% if total : 0.3525
% of quality: ': 0.8108747044917257

 val for kmean 2
f1_score : 0.0
accuracy_score : 1.0
size : 2
% if total : 0.0016666666666666668
% of quality: ': 0.0

 val for kmean 3
f1_score : 0.46022727272727265
accuracy_score : 0.4043887147335423
size : 319
% if total : 0.2658333333333333
% of quality: ': 0.7084639498432602

 val for kmean 4
f1_score : 0.512621359223301
accuracy_score : 0.4446902654867257
size : 452
% if total : 0.37666666666666665
% of quality: ': 0.5

/home/aura-alexis/github/ecg_qc_viz/env/lib64/python3.6/site-packages/scikit_learn-0.24.1-py3.6-linux-x86_64.egg/sklearn/metrics/_classification.py:1493: UndefinedMetricWarning:

F-score is ill-defined and being set to 0.0 due to no true nor predicted samples. Use `zero_division` parameter to control this behavior.


for record in df_train_sqi['record'].unique():
    print(f'record : {record}')
    print(df_train_sqi[df_train_sqi['record'] == record]['kmeans'].value_counts()/df_train_sqi[df_train_sqi['record'] == record].shape[0])

record : emg6+emg6- 77
1    0.518148
3    0.393333
2    0.083704
4    0.004444
0    0.000370
Name: kmeans, dtype: float64
record : ECG1+ECG1- 11
0    0.586667
3    0.205000
1    0.181667
2    0.026667
Name: kmeans, dtype: float64
record : EMG1+EMG1- 3
4    0.664762
2    0.139524
3    0.097619
1    0.097143
0    0.000952
Name: kmeans, dtype: float64


for kmean in df_train_sqi['kmeans'].unique():
    print(f'kmean : {kmean}')
    print(df_train_sqi[df_train_sqi['kmeans'] == kmean]['record'].value_counts()/df_train_sqi[df_train_sqi['kmeans'] == kmean].shape[0])

kmean : 3
emg6+emg6- 77    0.649144
ECG1+ECG1- 11    0.225550
EMG1+EMG1- 3     0.125306
Name: record, dtype: float64
kmean : 1
emg6+emg6- 77    0.724870
ECG1+ECG1- 11    0.169430
EMG1+EMG1- 3     0.105699
Name: record, dtype: float64
kmean : 2
EMG1+EMG1- 3     0.516755
emg6+emg6- 77    0.398589
ECG1+ECG1- 11    0.084656
Name: record, dtype: float64
kmean : 4
EMG1+EMG1- 3     0.991477
emg6+emg6- 77    0.008523
Name: record, dtype: float64
kmean : 0
ECG1+ECG1- 11    0.997167
EMG1+EMG1- 3     0.001889
emg6+emg6- 77    0.000944
Name: record, dtype: float64


df_train_sqi.groupby('record').count()


df_train_sqi

	timestamp_start	timestamp_end	qSQI_score	cSQI_score	sSQI_score	kSQI_score	pSQI_score	basSQI_score	annot_2	annot_3	record	consensus
0	2020-12-18 13:00:00.000000000	2020-12-18 13:00:02.000000000	1.000	0.501	1.924	6.664	0.499	0.967	1	1	emg6+emg6- 77	1
1	2020-12-18 13:00:02.000000000	2020-12-18 13:00:04.000000000	1.000	0.627	2.056	7.390	0.515	0.978	1	1	emg6+emg6- 77	1
2	2020-12-18 13:00:04.000000000	2020-12-18 13:00:06.000000000	0.857	0.820	1.333	3.494	0.484	0.975	1	1	emg6+emg6- 77	1
3	2020-12-18 13:00:06.000000000	2020-12-18 13:00:08.000000000	1.000	0.499	0.884	2.709	0.468	0.930	1	1	emg6+emg6- 77	1
4	2020-12-18 13:00:08.000000000	2020-12-18 13:00:10.000000000	1.000	0.482	1.817	5.480	0.433	0.991	1	1	emg6+emg6- 77	1

	timestamp_start	timestamp_end	qSQI_score	cSQI_score	sSQI_score	kSQI_score	pSQI_score	basSQI_score	classif	classif_avg
0	28800001	28802000	0.500	0.768	3.127	11.738	0.510	0.985	1	1.0
1	28802001	28804000	1.000	0.672	3.325	13.034	0.519	0.979	1	1.0
2	28804001	28806000	1.000	0.588	3.772	16.345	0.513	0.974	1	1.0
3	28806001	28808000	0.500	0.000	3.122	12.985	0.526	0.983	1	1.0
4	28808001	28810000	0.667	1.164	3.114	11.619	0.518	0.980	1	1.0

	qSQI_score	cSQI_score	sSQI_score	kSQI_score	pSQI_score	basSQI_score
count	6600.000000	6600.000000	6600.000000	6600.000000	6600.000000	6600.000000
mean	0.655783	0.675578	1.724129	9.954305	0.454387	0.949241
std	0.275371	0.164303	2.192340	12.758009	0.066733	0.052338
min	0.000000	0.000000	-2.995000	-1.851000	0.157000	0.421000
25%	0.500000	0.549000	-0.028000	0.780500	0.438000	0.933000
50%	0.667000	0.661000	1.290000	4.125500	0.465000	0.967000
75%	0.800000	0.795000	2.877000	12.846250	0.485000	0.983000
max	1.000000	1.409000	11.564000	143.106000	0.733000	1.000000

	qSQI_score	cSQI_score	sSQI_score	kSQI_score	pSQI_score	basSQI_score
count	1200.000000	1200.000000	1200.000000	1200.000000	1200.000000	1200.000000
mean	0.621078	0.668428	3.599284	15.732582	0.522501	0.977787
std	0.277671	0.311750	0.672979	4.511184	0.020708	0.014149
min	0.000000	0.000000	-0.373000	0.471000	0.434000	0.872000
25%	0.500000	0.551750	3.186500	12.228250	0.514000	0.971750
50%	0.667000	0.679000	3.647500	15.114000	0.525000	0.980000
75%	0.800000	0.856250	4.090250	19.560500	0.534000	0.987000
max	1.000000	1.404000	4.823000	26.217000	0.599000	0.999000

	qSQI_score	cSQI_score	sSQI_score	kSQI_score	pSQI_score	basSQI_score
count	5716.000000	5716.000000	5716.000000	5716.000000	5716.000000	5716.000000
mean	-0.013253	-0.073719	-1.756745	-4.779560	-0.069187	-0.027274
std	0.033159	-0.120096	1.539260	8.478286	0.050387	0.038090
min	0.000000	0.000000	-2.622000	-2.322000	-0.277000	-0.451000
25%	0.000000	-0.049500	-3.223750	-11.109250	-0.079000	-0.034000
50%	0.000000	-0.080000	-2.019500	-9.221000	-0.059000	-0.012000
75%	0.000000	-0.116250	-1.003250	-4.571500	-0.045750	-0.003000
max	0.000000	0.005000	6.801000	118.101000	0.134000	0.001000

0. Import & Initialisations¶

I. Random Forest¶

II. Decision Tree Classifer¶

Kmeans¶

Decision Tree¶

	qSQI_score	cSQI_score	sSQI_score	kSQI_score	pSQI_score	basSQI_score
count	5400.000000	5400.000000	5400.000000	5400.000000	5400.000000	5400.000000
mean	0.034704	0.007150	-1.875155	-5.778277	-0.068114	-0.028546
std	-0.002300	-0.147447	1.519361	8.246825	0.046024	0.038188
min	0.000000	0.000000	-2.622000	-2.322000	-0.277000	-0.451000
25%	0.000000	-0.002750	-3.214500	-11.447750	-0.076000	-0.038750
50%	0.000000	-0.018000	-2.357500	-10.988500	-0.060000	-0.013000
75%	0.000000	-0.061250	-1.213250	-6.714250	-0.049000	-0.004000
max	0.000000	0.005000	6.741000	116.889000	0.134000	0.001000

	qSQI_score	cSQI_score	sSQI_score	kSQI_score	pSQI_score	basSQI_score
count	6342.000000	6342.000000	6342.000000	6342.000000	6342.000000	6342.000000
mean	0.650863	0.676316	1.746178	10.184616	0.452987	0.950829
std	0.276360	0.164551	2.223179	12.921882	0.067158	0.051684
min	0.000000	0.000000	-2.995000	-1.851000	0.157000	0.421000
25%	0.500000	0.549250	-0.073500	0.779000	0.437000	0.938000
50%	0.667000	0.662000	1.306000	4.199000	0.465000	0.968000
75%	0.800000	0.796000	2.989750	13.771750	0.484000	0.984000
max	1.000000	1.409000	11.564000	143.106000	0.733000	1.000000

	qSQI_score	cSQI_score	sSQI_score	kSQI_score	pSQI_score	basSQI_score
count	626.000000	626.000000	626.000000	626.000000	626.000000	626.000000
mean	0.664117	0.750035	3.502923	14.964176	0.522174	0.978102
std	0.243201	0.284647	0.683919	4.443597	0.016771	0.013594
min	0.000000	0.000000	-0.373000	0.471000	0.434000	0.872000
25%	0.500000	0.598750	3.150250	11.888250	0.516000	0.972000
50%	0.667000	0.742000	3.325500	13.420000	0.524000	0.980000
75%	0.800000	0.912250	3.993000	18.343250	0.529750	0.987000
max	1.000000	1.404000	4.763000	25.005000	0.599000	0.999000

	qSQI_score	cSQI_score	sSQI_score	kSQI_score	pSQI_score	basSQI_score
count	-316.000000	-316.000000	-316.000000	-316.000000	-316.000000	-316.000000
mean	0.047958	0.080869	-0.118409	-0.998717	0.001073	-0.001272
std	-0.035459	-0.027351	-0.019899	-0.231460	-0.004363	0.000098
min	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000
25%	0.000000	0.046750	0.009250	-0.338500	0.003000	-0.004750
50%	0.000000	0.062000	-0.338000	-1.767500	-0.001000	-0.001000
75%	0.000000	0.055000	-0.210000	-2.142750	-0.003250	-0.001000
max	0.000000	0.000000	-0.060000	-1.212000	0.000000	0.000000

	qSQI_score	cSQI_score	sSQI_score	kSQI_score	pSQI_score	basSQI_score
count	258.000000	258.000000	258.000000	258.000000	258.000000	258.000000
mean	0.776709	0.657438	1.182143	4.292953	0.488802	0.910205
std	0.217680	0.157301	1.076263	5.107356	0.042770	0.053244
min	0.000000	0.000000	-2.266000	-1.371000	0.394000	0.710000
25%	0.667000	0.534000	0.494000	0.951500	0.464000	0.877500
50%	0.800000	0.639000	1.054000	2.867500	0.483000	0.920500
75%	1.000000	0.775250	1.781500	6.132500	0.508000	0.951500
max	1.000000	1.218000	5.563000	35.892000	0.699000	0.993000

	qSQI_score	cSQI_score	sSQI_score	kSQI_score	pSQI_score	basSQI_score
count	574.000000	574.000000	574.000000	574.000000	574.000000	574.000000
mean	0.574141	0.579429	3.704375	16.570601	0.522857	0.977443
std	0.304284	0.315877	0.645206	4.438230	0.024297	0.014736
min	0.000000	0.000000	0.490000	1.679000	0.439000	0.893000
25%	0.400000	0.506250	3.252250	13.500250	0.512000	0.970000
50%	0.571000	0.625000	3.819500	15.979500	0.529000	0.980000
75%	0.800000	0.781500	4.195250	20.645750	0.538000	0.988000
max	1.000000	1.397000	4.823000	26.217000	0.578000	0.999000

	timestamp_start	timestamp_end	qSQI_score	cSQI_score	sSQI_score	kSQI_score	pSQI_score	basSQI_score	annot_2	annot_3	consensus	rfc_pred	dtc_pred	kmeans
record
ECG1+ECG1- 11	1800	1800	1800	1800	1800	1800	1800	1800	1800	1800	1800	1800	1800	1800
EMG1+EMG1- 3	2100	2100	2100	2100	2100	2100	2100	2100	2100	2100	2100	2100	2100	2100
emg6+emg6- 77	2700	2700	2700	2700	2700	2700	2700	2700	2700	2700	2700	2700	2700	2700

	qSQI_score	cSQI_score	sSQI_score	kSQI_score	pSQI_score	basSQI_score
count	-316.000000	-316.000000	-316.000000	-316.000000	-316.000000	-316.000000
mean	0.202568	0.078009	-2.522231	-12.277648	-0.034055	-0.067237
std	-0.086604	-0.158576	0.431057	0.669127	0.018474	0.038509
min	0.000000	0.000000	-2.756000	-3.050000	-0.045000	-0.183000
25%	0.267000	0.027750	-2.758250	-12.548750	-0.048000	-0.092500
50%	0.229000	0.014000	-2.765500	-13.112000	-0.046000	-0.059500
75%	0.200000	-0.006250	-2.413750	-14.513250	-0.030000	-0.036500
max	0.000000	-0.179000	0.740000	9.675000	0.121000	-0.006000

	qSQI_score	cSQI_score	sSQI_score	kSQI_score	pSQI_score	basSQI_score
count	5716.000000	5716.000000	5716.000000	5716.000000	5716.000000	5716.000000
mean	-0.167864	-0.070860	0.647076	6.499370	-0.034059	0.038691
std	0.084304	0.011129	1.088304	7.577699	0.027551	-0.000321
min	0.000000	0.000000	0.134000	0.728000	-0.232000	-0.268000
25%	-0.267000	-0.030500	-0.456250	1.101000	-0.028000	0.053750
50%	-0.229000	-0.032000	0.408000	2.123500	-0.014000	0.046500
75%	-0.200000	-0.055000	1.200500	7.799000	-0.019000	0.032500
max	0.000000	0.184000	6.001000	107.214000	0.013000	0.007000

	timestamp_start	timestamp_end	qSQI_score	cSQI_score	sSQI_score	kSQI_score	pSQI_score	basSQI_score	classif	classif_avg	ecg_qc_pred	kmeans
2	28804001	28806000	1.000	0.588	3.772	16.345	0.513	0.967	1	1.0	1	0
32	28864001	28866000	0.667	1.403	3.267	13.217	0.436	0.927	1	1.0	1	0
52	28904001	28906000	1.000	0.000	3.183	12.400	0.439	0.979	1	1.0	1	0
54	28908001	28910000	1.000	0.673	3.682	15.837	0.506	0.950	1	1.0	1	0
62	28924001	28926000	0.800	0.618	3.944	17.471	0.507	0.948	1	1.0	1	0
...	...	...	...	...	...	...	...	...	...	...	...	...
1146	58692181	58694180	0.333	0.907	3.510	15.564	0.458	0.871	0	0.0	1	0
1172	58744181	58746180	0.571	0.486	2.639	8.788	0.487	0.985	0	0.0	1	0
1174	58748181	58750180	1.000	0.575	2.948	11.207	0.458	0.944	0	0.0	1	0
1176	58752181	58754180	1.000	0.572	4.072	17.966	0.453	0.987	0	0.0	1	0
1199	58798183	58799999	0.400	0.725	3.564	13.994	0.501	0.926	0	0.0	1	0